Sub-Connected Hybrid Precoding and Trajectory Optimization Using Deep Reinforcement Learning for Energy-Efficient Millimeter-Wave UAV Communications

Abstract

In this letter, a sub-connected hybrid precoding system was designed to realize energy-efficient millimeter-Wave (mmWave) unmanned aerial vehicle (UAV) communications. Considering the limited capacity of the UAV battery, the system was jointly optimized with a UAV trajectory to increase the energy efficiency of the UAV under quality-of-service (QoS) and power budget constraints. The dynamic motion of the UAV changes the channel condition between the UAV and terrestrial users over time. Hence, a joint optimization problem was formulated as a non-convex and time-sequential domain, solved using deep reinforcement learning (DRL). The performances of the proposed scheme and a fully-connected hybrid precoding scheme were compared in terms of energy efficiency and show higher results.